Forecast of Air Routes Market Share Based on Long Short-Term Memory Network
-
摘要: 当前主流航线市场的份额预测方法是服务质量指数(QSI)模型,但此方法需要模型线性化和大量人工经验. 提出基于长短时记忆网络的航线市场份额预测模型,利用该模型对航班市场份额进行预测,并通过在简化数据集上进行试验来验证模型的有效性. 以均方根误差为评价指标,对模型的参数进行优化,分别测试运力预测,QSI模型和提出的预测模型等3种方法的预测精度. 试验结果表明提出的模型能更好地预测航线市场份额,均方根误差在0.1左右.
-
关键词:
- 航空运输 /
- 市场份额预测 /
- 服务质量指数(QSI)模型 /
- 长短时记忆网络
Abstract: The mainstream forecast method of air route market share is quality of service index (QSI) model currently, but this method needs model linearization and a lot of manual experience. An airline market share forecasting model based on long short-term memory network was proposed. The model was used to forecast the flight market share, and the validity of the model was verified by experiments on simplified data sets. Taking root mean square error as the evaluation index, the parameters of the model were optimized, and the prediction accuracy of the three methods, such as capacity prediction, QSI model and the proposed prediction model, was tested respectively. Experimental results show that the proposed model can better predict the market share of airlines, and the root mean square error is about 0.1. -
表 1 数据集构成
Table 1. Data set composition
项目 月数 OD市场 / 个 航线 / 条 数据 / 组 训练集 11 80 218 813 测试集 3 74 203 211 表 2 数据集示例
Table 2. Example of data set
OD 市场 航空公司 当月航班 / 架 当月运载量 / 人 每航班座位 / 个 起飞城市市场份额 航线市场份额 上海—巴黎 法国航空 52 21876 421 0.021 0.500 上海—巴黎 中国国际航空 17 4505 265 0.060 0.158 上海—巴黎 东方航空 60 18840 314 0.311 0.343 表 3 模型参数调整
Table 3. Model parameter adjustment
批次大小 LSTM层数 / 层 隐藏节点 / 个 Dropout 是否为双向LSTM RMSE 1 1 3 0 是 0.125 1 2 3 0 是 0.102 1 3 3 0 是 0.103 1 4 3 0 是 0.107 1 2 2 0 是 0.136 1 2 4 0 是 0.109 4 2 3 0 是 0.147 8 2 3 0 是 0.152 1 2 3 0.1 是 0.096 1 2 3 0.2 是 0.099 1 2 3 0.1 否 0.125 表 4 QSI模型权重参数
Table 4. Weight parameters of QSI model
项目 是否加入起飞城市市场份额 当月航班数 当月运载量 每航班座位数 起飞城市市场份额 权重参数 是 0.584 0.306 0.101 0.009 否 0.590 0.307 0.103 — 表 5 QSI与LSTM模型结果对比
Table 5. Comparison between outcome of QSI and LSTM models
运力预测 是否加入起飞城市市场份额 QSI模型 LSTM模型 RMSE 0.168 是 0.152 0.096 否 0.157 0.129 表 6 2019年12月上海—大阪市场份额预测结果对比
Table 6. Forecast outcome of Shanghai-Osaka market in Dec 2019
航空公司 运力预测结果 QSI预测结果 LSTM预测结果 真值 HO 0.239 0.256 0.303 0.323 CA 0.289 0.280 0.219 0.231 CZ 0.169 0.162 0.160 0.149 MU 0.138 0.139 0.135 0.128 NH 0.095 0.093 0.091 0.084 JL 0.062 0.061 0.075 0.072 FM 0.020 0.009 0.016 0.013 -
[1] 陈蔚蔚, 许俐. 航空公司航线市场份额研究[J] . 中国民航学院学报,2006(2):30 − 33. [2] HALL M J, RAVINDRANATH R, BERMELL-GARCIA P, et al. Route networks within the air transport system: A comparative study of two European low-cost airlines using network metrics[C]//Proceedings of 2016 Annual IEEE Systems Conference (SysCon). Orlando: IEEE, 2016. [3] GRAF M, KIMMS A. Transfer price optimization for option-based airline alliance revenue management[J] . International Journal of Production Economics,2013,145(1):281 − 293. doi: 10.1016/j.ijpe.2013.04.049 [4] COLDREN G M, KOPPELMAN F S, KASTURIRANGAN K, et al. Modeling aggregate air-travel itinerary shares: Logit model development at a major US airline[J] . Journal of Air Transport Management,2003,9(6):361 − 369. doi: 10.1016/S0969-6997(03)00042-5 [5] 王斌, 张金隆. 航线流量组合预测模型研究[J] . 武汉理工大学学报(信息与管理工程版),2008,30(3):441 − 444. [6] KLOTH R J, EDSALL T J, FINE M, et al. Method and apparatus for implementing a quality of service policy in a data communications network: US6870812[P]. 2005-03-22. [7] 李纯青, 郝艳磊, 张军. 航空积分联盟及其对联盟合作伙伴的影响研究: 基于南航加入天合联盟的案例研究[J] . 管理案例研究与评论,2011,4(6):459 − 467. doi: 10.3969/j.issn.1674-1692.2011.06.005 [8] BARNHART C, SMITH B. Quantitative problem solving methods in the airline industry: A modeling methodology handbook[M]. Boston: Springer, 2012. [9] GROSCHE T, KLOPHAUS R, SEREDYŃSKI A. Market concentration in German air transport before and after the air Berlin bankruptcy[J] . Transport Policy,2020,94:78 − 88. doi: 10.1016/j.tranpol.2020.05.006 [10] 赵晓松, 朱金福, 葛伟. 联盟环境下国际航线市场份额预测方法研究[J] . 交通运输系统工程与信息,2018,18(2):33 − 39. [11] 陈娴, 朱金福, 刘月. 基于PSO-BP算法的联盟航线市场份额预测[J] . 华东交通大学学报,2020,37(1):61 − 69. [12] JÜRGEN S. Deep learning in neural networks: An overview[J] . Neural Networks,2015,61:85 − 117. doi: 10.1016/j.neunet.2014.09.003 [13] GRAVES A, MOHAMED A AND HINTON G. Speech recognition with deep recurrent neural networks[C]//Proceedings of IEEE International Conference on Acoustics, Speech and Signal Processing. Vancouver: IEEE, 2013. [14] HOCHREITER S, SCHMIDHUBER J. Long short-term memory[J] . Neural Computation,1997,9(8):1735 − 1780. doi: 10.1162/neco.1997.9.8.1735 [15] CHEN W H, ZHAO Z, LIU J M, et al. LSTM network: A deep learning approach for short-term traffic forecast[J] . IET Intelligent Transport Systems,2017,11(2):68 − 75. doi: 10.1049/iet-its.2016.0208 [16] SCHUSTER M, PALIWAL K K. Bidirectional recurrent neural networks[J] . IEEE Transactions on Signal Processing,1997,45(11):2673 − 2681. doi: 10.1109/78.650093 [17] 陈蔚蔚. 航线网络的经济性研究及其应用[D]. 南京: 南京航空航天大学, 2006.